feat(dashboard): optimize cache and AI pick confidence

services/ai_product_pick_agent.py

@@ -85,6 +85,8 @@ def _daily_sales_columns(conn) -> Dict[str, str]:
         "date": first_available(["snapshot_date", "日期", "訂單日期", "交易日期", "Date"]),
         "revenue": first_available(["總業績", "銷售金額", "業績", "金額", "Amount", "Sales", "Total"]),
         "qty": first_available(["數量", "銷售數量", "銷量", "Qty", "Quantity"]),
+        "profit": first_available(["毛利", "Profit", "利潤"]),
+        "cost": first_available(["總成本", "成本", "Cost", "進價"]),
     }
 
 
@@ -96,7 +98,7 @@ def _fetch_candidates(conn, limit: int) -> List[Dict[str, Any]]:
     from sqlalchemy import text
 
     sales_join = ""
-    sales_select = "0 AS sales_7d, 0 AS sales_prev_7d, 0 AS qty_7d"
+    sales_select = "0 AS sales_7d, 0 AS sales_prev_7d, 0 AS qty_7d, 0 AS profit_7d, 0 AS cost_7d"
     sales_cols = {}
     if _has_daily_sales_snapshot(conn):
         sales_cols = _daily_sales_columns(conn)
@@ -108,6 +110,10 @@ def _fetch_candidates(conn, limit: int) -> List[Dict[str, Any]]:
         date_col = _quote_identifier(sales_cols["date"])
         revenue_col = _quote_identifier(sales_cols["revenue"])
         qty_col = _quote_identifier(sales_cols["qty"])
+        profit_col = _quote_identifier(sales_cols["profit"]) if sales_cols.get("profit") else None
+        cost_col = _quote_identifier(sales_cols["cost"]) if sales_cols.get("cost") else None
+        profit_expr = f"COALESCE({profit_col}::numeric, 0)" if profit_col else "0"
+        cost_expr = f"COALESCE({cost_col}::numeric, 0)" if cost_col else "0"
         sales_join = """
             LEFT JOIN (
                 SELECT
@@ -118,7 +124,11 @@ def _fetch_candidates(conn, limit: int) -> List[Dict[str, Any]]:
                               AND {date_col}::date < CURRENT_DATE - 7
                              THEN COALESCE({revenue_col}::numeric, 0) ELSE 0 END) AS sales_prev_7d,
                     SUM(CASE WHEN {date_col}::date >= CURRENT_DATE - 7
-                             THEN COALESCE({qty_col}::numeric, 0) ELSE 0 END) AS qty_7d
+                             THEN COALESCE({qty_col}::numeric, 0) ELSE 0 END) AS qty_7d,
+                    SUM(CASE WHEN {date_col}::date >= CURRENT_DATE - 7
+                             THEN {profit_expr} ELSE 0 END) AS profit_7d,
+                    SUM(CASE WHEN {date_col}::date >= CURRENT_DATE - 7
+                             THEN {cost_expr} ELSE 0 END) AS cost_7d
                 FROM daily_sales_snapshot
                 GROUP BY {sku_col}
             ) sales ON sales.sku = lm.sku
@@ -127,11 +137,15 @@ def _fetch_candidates(conn, limit: int) -> List[Dict[str, Any]]:
             date_col=date_col,
             revenue_col=revenue_col,
             qty_col=qty_col,
+            profit_expr=profit_expr,
+            cost_expr=cost_expr,
         )
         sales_select = """
             COALESCE(sales.sales_7d, 0) AS sales_7d,
             COALESCE(sales.sales_prev_7d, 0) AS sales_prev_7d,
-            COALESCE(sales.qty_7d, 0) AS qty_7d
+            COALESCE(sales.qty_7d, 0) AS qty_7d,
+            COALESCE(sales.profit_7d, 0) AS profit_7d,
+            COALESCE(sales.cost_7d, 0) AS cost_7d
         """
 
     sql = text(f"""
@@ -236,7 +250,9 @@ def _fetch_candidates(conn, limit: int) -> List[Dict[str, Any]]:
                 NULL AS max_pchome_price,
                 0 AS sales_7d,
                 0 AS sales_prev_7d,
-                0 AS qty_7d
+                0 AS qty_7d,
+                0 AS profit_7d,
+                0 AS cost_7d
             FROM latest_momo lm
             JOIN competitor_prices cp
               ON cp.sku = lm.sku
@@ -257,24 +273,60 @@ def _score_candidate(row: Dict[str, Any]) -> Dict[str, Any]:
     sales_7d = _to_float(row.get("sales_7d"))
     sales_prev_7d = _to_float(row.get("sales_prev_7d"))
     qty_7d = _to_float(row.get("qty_7d"))
+    profit_7d = _to_float(row.get("profit_7d"))
+    cost_7d = _to_float(row.get("cost_7d"))
     history_points = int(_to_float(row.get("history_points")))
+    min_pchome_price = _to_float(row.get("min_pchome_price"))
     tags = _load_json_tags(row.get("tags"))
 
     gap_pct = ((momo_price - pchome_price) / pchome_price * 100) if pchome_price else 0
     sales_delta = ((sales_7d - sales_prev_7d) / sales_prev_7d * 100) if sales_prev_7d else None
+    if not profit_7d and cost_7d and sales_7d:
+        profit_7d = sales_7d - cost_7d
+    margin_rate = (profit_7d / sales_7d * 100) if sales_7d and profit_7d else None
 
-    price_score = max(0, min(38, gap_pct * 1.8 + 8))
-    match_component = max(0, min(24, match_score * 24))
+    price_score = max(0, min(40, gap_pct * 1.9 + 8))
+    match_component = max(0, min(30, match_score * 30))
     sales_component = 0
     if sales_7d > 0:
-        sales_component += min(10, sales_7d / 30000 * 10)
+        sales_component += min(9, sales_7d / 30000 * 9)
     if qty_7d > 0:
-        sales_component += min(5, qty_7d / 20 * 5)
+        sales_component += min(4, qty_7d / 20 * 4)
     if sales_delta is not None and sales_delta > 0:
-        sales_component += min(8, sales_delta / 40 * 8)
-    history_component = min(10, history_points * 2)
-    promo_component = 5 if any(tag in tags for tag in ["on_sale", "discount_10pct", "discount_20pct", "discount_30pct"]) else 0
-    score = round(min(100, price_score + match_component + sales_component + history_component + promo_component), 1)
+        sales_component += min(7, sales_delta / 40 * 7)
+    margin_component = 0
+    if margin_rate is not None:
+        margin_component = max(0, min(10, margin_rate / 35 * 10))
+    history_component = min(12, history_points * 2.4)
+    promo_component = 0
+    if any(tag in tags for tag in ["on_sale", "discount_10pct", "discount_20pct", "discount_30pct"]):
+        promo_component += 5
+    if "high_rating" in tags:
+        promo_component += 3
+    if "low_stock" in tags:
+        promo_component -= 4
+    price_position_component = 0
+    if min_pchome_price and pchome_price:
+        if pchome_price <= min_pchome_price * 1.03:
+            price_position_component = 6
+        elif pchome_price <= min_pchome_price * 1.08:
+            price_position_component = 3
+
+    opportunity_score = min(
+        100,
+        price_score + sales_component + margin_component + promo_component + price_position_component,
+    )
+    evidence_quality = min(
+        100,
+        match_component
+        + history_component
+        + (12 if sales_7d > 0 else 0)
+        + (8 if margin_rate is not None else 0)
+        + (8 if row.get("competitor_product_id") and row.get("competitor_product_name") else 0)
+        + (6 if row.get("crawled_at") else 0),
+    )
+    score = round(min(100, opportunity_score + evidence_quality * 0.35), 1)
+    confidence = round(max(0.45, min(0.98, (score * 0.65 + evidence_quality * 0.35) / 100)), 3)
 
     if gap_pct >= 10:
         angle = "PChome 價格優勢明顯"
@@ -292,15 +344,26 @@ def _score_candidate(row: Dict[str, Any]) -> Dict[str, Any]:
     ]
     if sales_7d > 0:
         reason_parts.append(f"近 7 天銷售額 ${sales_7d:,.0f}")
+    if margin_rate is not None:
+        reason_parts.append(f"近 7 天毛利率 {margin_rate:.1f}%")
     if history_points:
         reason_parts.append(f"已有 {history_points} 筆 PChome 歷史快照")
+    if price_position_component:
+        reason_parts.append("目前 PChome 價格接近 30 天低點")
+    if "high_rating" in tags:
+        reason_parts.append("PChome 商品評價訊號佳")
+    if "low_stock" in tags:
+        reason_parts.append("PChome 庫存偏低，需留意供貨")
 
     return {
         **row,
         "gap_pct": round(gap_pct, 1),
         "sales_7d_delta": round(sales_delta, 1) if sales_delta is not None else 0,
         "pick_score": score,
-        "confidence": round(max(0.45, min(0.98, score / 100)), 3),
+        "confidence": confidence,
+        "evidence_quality": round(evidence_quality, 1),
+        "opportunity_score": round(opportunity_score, 1),
+        "margin_rate": round(margin_rate, 1) if margin_rate is not None else None,
         "reason": "；".join(reason_parts),
     }
 
@@ -315,6 +378,9 @@ def _write_pick(conn, pick: Dict[str, Any]) -> None:
             "generated_at": datetime.now().isoformat(timespec="seconds"),
             "inputs": ["products", "price_records", "competitor_prices", "competitor_price_history", "daily_sales_snapshot"],
             "score": pick["pick_score"],
+            "opportunity_score": pick.get("opportunity_score"),
+            "evidence_quality": pick.get("evidence_quality"),
+            "margin_rate": pick.get("margin_rate"),
         },
         "competitor": {
             "source": "pchome",

services/cache_manager.py

@@ -7,7 +7,9 @@ ADR-017 Phase 3f-2: 將 sales/import/export/daily 會共同碰到的
 module-level cache 收斂到這裡，避免各 route 各自持有一份 dict。
 """
 
+import os
 import time
+from pathlib import Path
 
 
 class FingerprintCache:
@@ -59,6 +61,8 @@ _DASHBOARD_DATA_CACHE = {
     'full_timestamp': None,
 }
 _DASHBOARD_CACHE_TTL = 1800
+_BASE_DIR = Path(__file__).resolve().parents[1]
+_DASHBOARD_SHARED_CACHE_FILE = _BASE_DIR / "data" / "dashboard_full_cache.pkl"
 
 
 def cleanup_sales_cache():
@@ -123,3 +127,9 @@ def clear_dashboard_cache():
         'full_data': None,
         'full_timestamp': None,
     })
+    try:
+        os.remove(_DASHBOARD_SHARED_CACHE_FILE)
+    except FileNotFoundError:
+        pass
+    except OSError:
+        pass